Multiple Logistic Regression: Types of Stress and Outcome of In Vitro Fertilization Treatment in Infertile Women

Author

Obehi Winnifred Ikpea

Published

August 29, 2024

1 Introduction

According to the World Health Organization, estimates suggest that approximately one in every six people of reproductive age worldwide experiences infertility in their lifetime. Infertility is a disease of the male or female reproductive system defined by the failure to achieve a pregnancy after 12 months or more of regular unprotected sexual intercourse. This disease is caused by a number of different factors in either of the reproductive systems. In addition to these causes, lifestyle factors like smoking, excessive alcohol intake and obesity, and exposure to environmental pollutants and toxins can affect fertility (World Health Organization, 2020). There are different treatments for infertility; an example is the In Vitro Fertilization (IVF), a complex series of procedures that leads to pregnancy. The IVF procedure is the most effective type of fertility treatment that involves the handling of eggs or embryos and sperm. Although the treatment is effective, it raises the chances of certain health problems like stress, as the treatment can be draining for the body, mind, and finances (Mayo Clinic Staff, 2023).

Stress is the state of worry or mental tension caused by a difficult situation. It is a natural human response that prompts us to address challenges and threats in our lives (World Health Organization, 2023). Stress can be categorized into acute stress, chronic stress, and perceived stress. Acute stress is short-term stress that comes and goes quickly, and can be positive or negative. Acute stress can be measured using the State-Trait Anxiety Inventory, a commonly used measure of trait and state anxiety developed in 1970 to diagnose anxiety and distinguish it from depressive syndromes. High scores indicate high anxiety levels, and low scores indicate low anxiety levels (American Psychological Association, 2011) (ScienceDirect Topics, 2024). Chronic stress is long-term stress that goes on for a long duration of time (Cleveland Clinic, 2024). Chronic stress can be measured by the Social Readjustment Rating Scale, a tool developed in 1967 to measure the amount of stress a person has experienced recently (Mind Tools Content Team, 2023). It is an inventory of the most common life stressors but can be difficult to interpret due to differences in individual coping abilities (Brandeis University, n.d.).

Perceived stress refers to the feelings or thoughts that an individual has about how much stress they are under at a given point in time or over a given time period (Phillips, 2013). Perceived stress is measured by the Perceived Stress Scale, the most widely used psychological instrument for measuring the perception of stress. It is a measure of the degree to which situations in one’s life are appraised as stressful (Cohen, 1994). Stress affects the body physically with symptoms like chest pain, headaches, weakened immune system, and high blood pressure; psychologically with symptoms like anxiety, depression, and panic attacks; and behaviorally with symptoms like smoking, overeating, and alcohol use disorder (Cleveland Clinic, 2024).

Studies show that acute

Aim

The aim of this project is to investigate the association between specific types of stress (Acute, Chronic, and Perceived) and the outcome of pregnancy after IVF for a population of infertile women, while accounting for age and history of mental illness as potential confounders. Through this project, we would like to gain insight into how psychological factors like stress can influence physical outcomes in reproductive health.

Scientific Objectives

The following scientific objectives will be considered:

  • To investigate the association between pregnancy outcome after IVF and chronic stress, measured by the Social Readjustment Rating Scale (SRRS), adjusting for age , and history of mental illness.

  • To investigate the association between pregnancy outcome after IVF and acute stress, measured by the Spielberger Trait-State Anxiety Inventory (STAI), adjusting for age, and history of mental illness.

  • To investigate the association between pregnancy outcome after IVF and perceived stress, measured by the Perceived Stress Scale (PSS), adjusting for age, and history of mental illness.

2 Data Description

The data comes from a prospective study examining the relationship between various types of stress and the outcome of In Vitro Fertilization (IVF) Treatment in Infertile Women. It comprises of 172 participants and 12 measurements, namely:

  1. Binary indicators for:

    • Successful pregnancy

    • Cause of infertility

    • History of successful implantation

    • History of abortion

    • History of mental illness

    • History of thyroid problems and diabetes

  2. Stress measurements:

    • Chronic stress: measured by the Social Readjustment Rating Scale (SRRS)

    • Acute stress: measured by the Spielberger State-Trait Anxiety Inventory (STAI)

    • Perceived stress: measured by the Perceived Stress Scale (PSS)

  3. Demographic and medical history:

    • Age (in years)

    • Education (in years)

    • Infertility period (in years)

3 Methods

Data Preprocessing

The dataset was tidied to ensure completeness and correctness before multiple analyses were implemented. This process involved:

  • Recoding the values of the categorical variables to their specific labels and changing the data types of specific variables.

Exploratory Data Analysis

Exploratory analysis was implemented on the dataset to identify errors and anomalies, understand the distribution of the variables, and gain visual and numeric insights into the associations between the outcome and predictor variables. This method was carried out in various stages:

  • A summary statistics table was generated to present summary measures for the variables in the dataset used for the analyses.

  • Histograms were created to illustrate the distribution of the outcomes visually.

  • Stratified box-plots were created to visualize the distribution of the measures of the different types of stress (Chronic, Acute, Perceived) across the different pregnancy outcomes. The following observations were made:

Summary Statistics Table

Data Visualizations

Statistical Analysis

The aim of this project was to investigate the associations between pregnancy outcomes after IVF and three types of stress for a population of infertile women, adjusting for age and history of mental illness. To explore this aim, we proposed three scientific objectives. For each objective, we fitted multiple binary logistic regression models of the form:

\(Log(odds(Y_i))=\beta_0 + \beta_1X_{i1} + \beta_2Age + \beta_3I_{History=1}\)

Where:

  • \(Y_i\) is the outcome of interest, a binary variable with values of 1 representing successful pregnancy, and 0 representing unsuccessful pregnancy.

  • \(X_{i1}\) is the predictor of interest, and for each objective represents the measure of the respective stress type.

  • The parameters \(\beta_0, \beta_1, \beta_2, \beta_3\) are unknown and estimated by the method of maximum likelihood.

  • \(\beta_1\) is the parameter of interest and is interpreted as the difference in log odds of successful pregnancy between two subpopulations of infertile women of the same age and with the same mental illness history that differ in their respective stress measure by 1 unit. The estimated parameter will be interpreted on the odds ratio scale.

  • \(I_{History=1}\) is an indicator variable for history of mental illness, where 1 represents presence of history and 0 represents absence.

  • We evaluated statistical evidence by conducting multiple hypothesis tests using the Wald test. For each objective, we tested the following hypotheses:

    • \(H_0: \beta_1 = 0\) (Null hypothesis: There is no association between pregnancy outcome and the stress type, adjusting for age and history of mental illness.)

    • \(H_1: \beta_1 \neq 0\) (Alternative hypothesis: There is an association between pregnancy outcome and the stress type, adjusting for age and history of mental illness.)

  • To avoid overfitting and ensure generalizability, we limited the number of predictors in the model. Using the rule of thumb to have no more than \(n * p_{min}/15\) predictors, where \(n\) is the sample size and \(p_{min}\) is the proportion of the less frequent outcome. Hence, our model had 3 predictor variables with the stress type being the predictor of interest for each of the models.

Model Diagnostics

The validity of regression coefficient estimates, confidence intervals, and significance tests relies on these assumptions: binary outcomes, independence of observations, large sample size, and linearity in the logit. Outliers, influential observations, and multicollinearity were also investigated as these influence the regression coefficients and validity of the model.

  • Binary outcomes: This assumption was met as our outcome variable (pregnancy success) was dichotomous.

  • Independence of observations: This assumption was satisfied as there were no repeated measurements for the participants in the study.

  • Large sample size: Logistic regression requires a large sample size, with a general rule of thumb being a minimum of 10 cases per predictor variable. Our analysis met this assumption, as there were more than 50 cases per variable.

  • Linearity in the logit: This assumption requires the predictor variables to be linearly related to the log odds of the response variable. We assessed this using marginal model plots, where the response variable is plotted against each predictor variable.

  • Outliers and influential observations: We assessed these in the diagnostic checking section, performing tests for outliers and using various influence measures such as studentized residuals, Cook’s distance and leverage statistics.

  • Multicollinearity: This assumptions requires the predictor variables to not be highly correlated with each other, and was diagnosed with the variance inflation factor.

4 Results

Exploratory Data Analysis

Summary Statistics Table

Table 1. Summary Statistics
Variable N = 1721
pregnancy
    Successful 124, 72% 0
    Unsucessful 48, 28% 0
Acute_stress
    Mean, Median, SD 39, 39, 9
     Range 20, 62
    (25% 75%) (32 45)
Chronic_Stress
    Mean, Median, SD 145, 133, 109
     Range 0, 449
    (25% 75%) (63 199)
Perceived_stress
    Mean, Median, SD 15, 15, 7
     Range 0, 39
    (25% 75%) (10 19)
mental_illness
    History 7, 4.1% 0
    No History 165, 96% 0
Age
    Mean, Median, SD 31.4, 31.0, 4.7
     Range 20.0, 45.0
    (25% 75%) (28.0 35.0)
1 n, % N missing

Histograms

Histogram for Chronic Stress

Histogram for Acute Stress

Histogram for Percieved Stress

Boxplots

Chronic stress and Pregnancy outcome

Acute stress and Pregnancy outcome

Percieved stress and Pregnancy outcome

Fitted Models

Logisitc Regression Model Results

Model 1: Chronic Stress

  pregnancy
Predictors Odds Ratios CI p
(Intercept) 2.29 0.23 – 23.28 0.477
Chronic Stress 1.00 1.00 – 1.00 0.837
Age 1.00 0.93 – 1.08 0.929
mental illness [1] 0.49 0.10 – 2.60 0.365
Observations 172
R2 Tjur 0.005

Model 2: Acute Stress

  pregnancy
Predictors Odds Ratios CI p
(Intercept) 1.16 0.09 – 15.21 0.908
Acute stress 1.02 0.99 – 1.06 0.235
Age 1.00 0.93 – 1.08 0.982
mental illness [1] 0.42 0.09 – 2.27 0.282
Observations 172
R2 Tjur 0.013

Model 3: Percieved Stress

  pregnancy
Predictors Odds Ratios CI p
(Intercept) 0.95 0.08 – 11.14 0.968
Perceived stress 1.05 1.00 – 1.11 0.048
Age 1.01 0.94 – 1.09 0.803
mental illness [1] 0.34 0.07 – 1.89 0.192
Observations 172
R2 Tjur 0.029

5 Discussion

This project aimed to investigate the associations between specific types of stress (Acute, Chronic, and Perceived) and the outcome of pregnancy after In Vitro fertilization treatment for a population of infertile women. The data utilized for this project was obtained from a prospective study with the same objective. Descriptive statistics revealed that the participants ranged from 20 to 45 years old (Mean=31.4, SD=7). Their score on the:

  • Chronic Stress Measure: The Social Readjustment Rating Scale (SRRS) ranged from 0 to 499 (Mean=145, SD=109). Seventy-five percent of the participants had SRRS scores of 199 or lower, indicating that most experienced a moderate to the low-stress level.

  • Acute Stress Measure: The Spielberger State-Trait Anxiety Inventory (STAI) ranged from 20 to 62 (Mean=39, SD=9). Seventy-five percent of the participants had STAI scores of 45 or lower, indicating no or low to moderate anxiety.

  • Perceived Stress Measure: The Perceived Stress Scale (PSS) ranged from 0 to 39 (Mean=15, SD=7). Seventy-five percent of the participants had PSS scores of 19 or lower, indicating low to moderate perceived stress levels.

The majority of the participants had no history of mental illness (96%) and had a successful pregnancy (72%).

Our analysis of the stratified boxplots, which show the distribution of stress measures for each pregnancy outcome, revealed some interesting findings. While the median SRRS and STAI scores were similar for successful and unsuccessful pregnancies, the PSS median score was notably higher for successful pregnancies than unsuccessful ones.

We fitted three binary logistic regression models to explore the objectives in greater detail. To investigate these associations, we estimated the odds of a successful pregnancy given the predictor of interest for each model, adjusting for potential confounders. After adjusting for Age and the history of mental illness, our findings revealed that:

  • Perceived stress was significantly positively associated with pregnancy outcome (Adjusted Odds Ratio [AOR] = 1.05; 95% Confidence Interval [CI]: 1.00-1.11; p =0.05). Participants with perceived stress measured by the Perceived Stress Scale (PSS) had 1.05 times the odds of a successful pregnancy.

  • The Chronic and Acute Stress models yielded non-significant findings for these predictors of interest.

We conducted diagnostic tests to check for violations of model assumptions and found that our models did not violate the absence of multicollinearity or independence assumptions. There were no highly influential observations, although some observations had high leverage. However, we identified a violation of the Linearity assumption for Model 1 (Chronic Stress as Predictor of Interest) and Model 3 (Perceived Stress as Predictor of Interest). To address this violation, we applied non-linear transformations:

  • For Model 1, we included a quadratic term for the Chronic Stress and Age variables.

  • For Model 3, we fitted a natural spline with two degrees of freedom for the Perceived Stress variable.

We implemented Analysis of Variance (ANOVA) to assess if adding the quadratic term to Model 1 and the natural spline to Model 3 improved the fit. From the results, we concluded that:

  • Adding the quadratic term to the Age and Chronic Stress variables for Model 1 did improve the fit (Pr > Chi = 0.045).

  • Adding the natural splines to the Perceived Stress variable did not improve the model fit (Pr > Chi = 0.5).

6 Appendix

Multicollinearity

The following table presents the Variance Inflation Factors (VIF) for the models:

           Chronic_Stress                       Age as.factor(mental_illness) 
                 1.055338                  1.019051                  1.036599 
             Acute_stress                       Age as.factor(mental_illness) 
                 1.050782                  1.015362                  1.035675 
         Perceived_stress                       Age as.factor(mental_illness) 
                 1.068320                  1.003623                  1.064650 

Influence Index Plots

Influence index plots for all models are displayed in the layout below:

Model 1

Figure 1: Influence Index Plot for Model 1

Model 2

Figure 2: Influence Index Plot for Model 2

Model 3

Figure 3: Influence Index Plot for Model 3

CR Plots for Linearity

Model 1

Figure 4: CR Plot for Model 1

Model 2

Figure 5: CR Plot for Model 2

Model 3

Figure 6: CR Plot for Model 3

Outliers

The outlier tests for each model are summarized below:

No Studentized residuals with Bonferroni p < 0.05
Largest |rstudent|:
    rstudent unadjusted p-value Bonferroni p
23 -1.667524            0.09541           NA
No Studentized residuals with Bonferroni p < 0.05
Largest |rstudent|:
     rstudent unadjusted p-value Bonferroni p
172 -1.810794           0.070173           NA
No Studentized residuals with Bonferroni p < 0.05
Largest |rstudent|:
     rstudent unadjusted p-value Bonferroni p
165 -2.083486           0.037207           NA

Sensitivity Analyses

Model A

Fitted models

  pregnancy
Predictors Odds Ratios CI p
(Intercept) 392917.62 0.32 – 11981556264637.58 0.102
Chronic Stress 0.99 0.98 – 1.00 0.091
Chronic Stress^2 1.00 1.00 – 1.00 0.068
Age 0.47 0.16 – 1.16 0.137
Age^2 1.01 1.00 – 1.03 0.132
mental illness [1] 0.57 0.11 – 3.16 0.497
Observations 172
R2 Tjur 0.034

Diagnostic Checking

Multicollinearity
           Chronic_Stress       I(Chronic_Stress^2)                       Age 
                 8.530253                  8.440482                152.386687 
                 I(Age^2) as.factor(mental_illness) 
               152.408622                  1.046963 
Influence Index Plots

::: {layout-ncol=“2”} ::: column

Figure 7: Influence Index Plot for Model A

:::

CR plots for linearity
Figure 8: CR Plot for Model A
Outliers

The outlier tests for each model are summarized below:

No Studentized residuals with Bonferroni p < 0.05
Largest |rstudent|:
    rstudent unadjusted p-value Bonferroni p
34 -1.988171           0.046793           NA

Model B: Percieved Stress

Fitted models

  pregnancy
Predictors Odds Ratios CI p
(Intercept) 0.64 0.05 – 8.67 0.737
Perceived stress [sqrt] 1.39 1.02 – 1.92 0.039
Age 1.01 0.94 – 1.08 0.837
mental illness [1] 0.37 0.07 – 1.98 0.211
Observations 172
R2 Tjur 0.031

Diagnostic Checking

Multicollinearity
   sqrt(Perceived_stress)                       Age as.factor(mental_illness) 
                 1.043132                  1.001658                  1.041443 
Influence Index Plots
Figure 9: Influence Index Plot for Model B
CR plots for linearity
Figure 10: CR Plot for Model B
Outliers

The outlier tests for each model are summarized below:

No Studentized residuals with Bonferroni p < 0.05
Largest |rstudent|:
     rstudent unadjusted p-value Bonferroni p
165 -1.950783           0.051083           NA

Anova Results

Analysis of Deviance Table

Model 1: pregnancy ~ Chronic_Stress + Age + as.factor(mental_illness)
Model 2: pregnancy ~ Chronic_Stress + I(Chronic_Stress^2) + Age + I(Age^2) + 
    as.factor(mental_illness)
  Resid. Df Resid. Dev Df Deviance Pr(>Chi)  
1       168     202.87                       
2       166     196.66  2   6.2104  0.04482 *
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Analysis of Deviance Table

Model 1: pregnancy ~ Perceived_stress + Age + as.factor(mental_illness)
Model 2: pregnancy ~ sqrt(Perceived_stress) + Age + as.factor(mental_illness)
  Resid. Df Resid. Dev Df Deviance Pr(>Chi)
1       168     198.83                     
2       168     198.61  0  0.22532         

References

American Psychological Association. (2011). The state-trait anxiety inventory (STAI). American Psychological Association. https://www.apa.org/pi/about/publications/caregivers/practice-settings/assessment/tools/trait-state
Brandeis University. (n.d.). Social readjustment rating scale (SRRS). https://www.brandeis.edu/roybal/docs/Social%20Readjustment%20rating%20scale_Website.pdf
Cleveland Clinic. (2024). What is stress? Cleveland Clinic. https://my.clevelandclinic.org/health/diseases/11874-stress
Cohen, S. (1994). Perceived stress scale. https://www.mindgarden.com/documents/PerceivedStressScale.pdf
Mayo Clinic Staff. (2023). In vitro fertilization (IVF). Mayo Clinic; Mayo Foundation for Medical Education; Research. https://www.mayoclinic.org/tests-procedures/in-vitro-fertilization/about/pac-20384716
Mind Tools Content Team. (2023). The holmes and rahe stress scale. Mind Tools. https://www.mindtools.com/avn893g/the-holmes-and-rahe-stress-scale
Phillips, A. C. (2013). Perceived stress. Encyclopedia of Behavioral Medicine, 1453–1454. https://doi.org/10.1007/978-1-4419-1005-9_479
ScienceDirect Topics. (2024). State trait anxiety inventory - an overview. In ScienceDirect. https://www.sciencedirect.com/topics/medicine-and-dentistry/state-trait-anxiety-inventory
World Health Organization. (2020). Infertility. World Health Organization. https://www.who.int/news-room/fact-sheets/detail/infertility
World Health Organization. (2023). Stress. World Health Organization. https://www.who.int/news-room/questions-and-answers/item/stress